From CRUD to Autonomous Agents: Zero-Trust Security Architecture for AI-Native Enterprise Systems

arXiv·high signalXBlueskyLinkedInCopy link

Formalizes the security gap between traditional CRUD/REST architectures and AI-native systems where LLMs act as cognitive orchestrators. Proposes a zero-trust semantic gateway framework with formal validation for enterprise deployments of autonomous agents. The core tension: probabilistic LLM outputs must pass through deterministic security boundaries. Introduces formal methods for validating agent actions before they reach enterprise resources — directly applicable to anyone building agent-based internal tools that touch production data or APIs.